Vous êtes sur la page 1sur 5

Development Lightweight Design

FE Simulation of Honeycomb Core


Sandwich Panels for the Body
Lightweight honeycomb sandwich structures have been increasingly employed in the automotive
industry: from parcel-shelf to load-floor applications. There can be an infinite variety of possible
solutions adopted for these parts depending on the choice of the core geometries and of the materi-
als. In the design phase an efficient prediction tool is needed Rieter has developed a mathematical
model based on a multi-scale asymptotic technique.

26 ATZ 01I2008 Volume 110


Author

1 Introduction the static or dynamic behaviour) espe- Davide Caprioli,


cially if these models have to be embed- MSc PhD Mechanical
Two main factors drive the progressive ded for example inside a complete vehi- Engineering, is Head
introduction of sandwich structures in cle model. of Vehicle and Compo-
the automotive industry: value of stiff- nent Simulation,
ness to weight ratio and, not surprising- Centre of Excellence
ly, material and processing cost. The 2 Proposed Procedure for Vehicle Acoustics,
problem of weight reduction, common at Rieter Automotive
to all forms of transportation for reasons The method essentially proposed here Systems in Winterthur
of fuel economy and driving perform- consists of a homogenization process (Switzerland).
ance, demands solutions that maximize and goes through the following steps:
structural stiffness with less material. In identifying the different scales character-
the automotive industry, sandwiches izing the physical problem; studying the
have already made their first steps in re- influence of each scale on the others; de-
placing conventional solutions for both veloping a technique to model this influ-
non-structural components such as ence. Once the homogenized process is
parcel shelf or load floor panels in most developed, it is possible to analyze the
of the automotive market, and with dynamics of the largest scale without
structural function in high performance considering the other scales.
sports models. Rieter has succeeded in extending
Certainly, the versatility of composite the validity of the two-scales perturba-
materials with cell structure, Figure 1, is tion asymptotic technique, developed in
revolutionizing the architecture of recent literature [1], to determine the
sandwich structures, and designs with constitutive dynamic equations of the
multiple choices of periodic spatial core equivalent homogenous medium repre-
arrangements are becoming a reality. senting the honeycomb cellular struc-
Based on demands of mechanical behav- ture: the outcome of this process is the
iour, cost, weight and acoustic perform- determination, for any honeycomb core
ance, the design can be rationally opti- geometry, of all the nine compliance
mized when certain theoretical guid- matrix terms of the equivalent homog-
ance is available; however the availabil- enous orthotropic material represent-
ity of analytical solutions appears to be ing the considered honeycomb cellular
lagging. It must also be realized that structure.
even with current high-speed comput- The derivations here described start
ers, it is still difficult to tackle a full de- from the general partial differential
scription of periodic sandwich struc- equations that represent the dynamic be-
tures by employing Finite Elements haviour of the honeycomb structure
commercial codes: the number of de- (Cauchys equations) and re-formulate
grees of freedom necessary to build up a them in a weak form, which is as a vari-
detailed model of the desired panel ational problem. This type of approach is
make its simulation very expensive (of made necessary by the fact that, in any

Figure 1: Comparison (FE modelling) of honeycomb validation model (left) to homogenized


model (right) of a sandwich structure

ATZ 01I2008 Volume 110 27


Development Lightweight Design

Figure 2: Static deformation comparison


(validation model vs. homogenized model)

honeycomb structure, the transition equation while the second is the macro- results obtained for a typical honeycomb
from the cell wall to the air around it de- scale equation. core structure with hexagonal cell (with
termines discontinuities in the local den- The macro-scale equation provides the 4 mm side length and 8 mm width) are
sity and in the local elasticity tensor of solution of the equivalent material iden- considered. In particular considering a
the honeycomb structure. Therefore one tification, since it gives the definition of honeycomb sandwich plate (1000 mm
considers a 3D-periodic body occupying the averaged density and stiffness tensor, 900 mm) with 1 mm aluminium skins,
a bounded region in the Euclidean R3 which are the equivalent properties and 2 mm core height made with poly-
space, defined by coordinates x1, x2 und needed to be identified. Indeed analyz- propylene.
x3. Assuming a linear behaviour, the ing the expression of the average stiff- The validation of the multi-scale tech-
equations that describe the dynamics of ness tensor, it can be seen that together nique is carried out employing the com-
the body can be written as follows: with known properties such as the vol- mercial FE code M.S.C. Nastran. The vali-
ume of the 3D periodic cell geometry, dation process consists in the compari-
and the stiffness tensor of the material son between the dynamic behaviours of
Dv
___

Dti ___
j
( ul
x Cijkl ___ )
x = fi in 
k
Eq. (1) constituting the honeycomb, the expres-
sion presents a third order tensor qkl,
two different finite element models of a
sandwich structure with honeycomb
Here u is the displacement field, v is the which is unknown. core, Figure 1:
D
velocity field, __
Dt is the material deriva- The identification of this unknown the validation model, where the thin-
tive, is the density, Cijkl are the terms of micro-scale tensor comes from the varia- walled core of the selected honeycomb
the elastic tensor, and fi are the terms tional equation system expressing the is described with a fine mesh made of
relative to the mass loads. For a linear micro-scale problem. This equation sys- linear plate elements (CQUAD4), to
elastic body the tensor Cijkl must be posi- tem can be solved by means of numerical which the polypropylene material
tive definite and must satisfy the follow- computation on the cell domain. And properties attributed. This model end-
ing symmetry conditions: thanks to the symmetry of the equiva- ed up having nearly 218.000 nodes.
lent elastic tensor only the solution of the homogenized model, where the
Cijkl = Cjikl = Cijlk = Cklij Eq. (2) ten terms of the unknown micro-scale volume bounding the core layer has
tensor kl
(y1, y2, y3) have to be performed.
q
been meshed with 3D elements. This
For any periodic structure the elasticity Once all the qkl have been obtained by involved a much lower modelling ef-
tensor Cijkl and the density will be peri- numerical solution on the cell domain, fort compared to the validation mod-
odic functions of the spatial coordinates. for the computation of the equivalent el, leading to an FE description with
As a consequence of this, it will be elastic tensor we can directly apply the only approximately 5500 nodes. To
possible to isolate a unit cell of the global macro-scale dynamic equation already these brick elements (CHEXA) the or-
structure, intended as the region that obtained from the endless variational thotropic material (MAT 9) behaviour
covers a single spatial period of the elas- equations deriving from Eq. (1). with the equivalent homogeneous
ticity tensor and of the density. By defin- core properties have been applied.
ing, on this domain, local coordinates Both the FE models present two skins,
(micro-scale coordinates) (y1, y2, y3) one 3 Numerical/Numerical Validation which are simulated via isotropic linear
has u = u (x1, x2, x3, y1, y2, y3, t), v = v (x1, x2, plate elements (CQUAD4). A first com-
x3, y1, y2, y3, t) and C = C (y1, y2, y3). The To confirm the results of the proposed parison between the two mentioned nu-
micro-scale coordinates are able to de- formulation, we compared, for the most merical models was performed with re-
scribe the small oscillating perturbations common honeycomb core configura- spect to a static solution, loading the
of the solution, while the macro-scale co- tions, a detailed FE model versus our ho- plate with a uniform load of 10 N on a
ordinates are able to describe the average mogenized approach. With this type of central square area, and with simple
spatial trend of the solution. validation, since no problem regarding support condition on two opposite pan-
Using the periodicity of the perturba- the material and geometry uncertainty el sides, on the skin opposite to the load.
tions and the Green theorem, two of the arises, between the compared models, it The solution of the validation model
endless variational equations are extract- is possible to better appreciate the accu- took 19 min while the solution of the
ed: the first represents the micro-scale racy of the proposed approach. Here the homogenized model took approximate-

28 ATZ 01I2008 Volume 110


4 Numerical/Experimental Validation

Similar results as mentioned in Chapter3


have been obtained with numerical vali-
dation of other core shapes, and with
numerical and experimental validation
of flat sandwich plates [2]. Here for brev-
ity the application of the proposed ho-
mogenization method to the simulation
of a load floor panel in the trunk will be
reported, Figure 4.
This load floor is composed of sand-
wich carrier material, named RHOC
(Rieter Honeycomb), which is made up
of the following starting layers: two skin
layers made out of EAC material (glass
fibre based material), a polypropylene
tubular honeycomb core (17 mm thick-
ness), and on top a non woven carpet. In
Figure 3: Root mean square (RMS) of mobility comparison (validation model vs. the production process, this starting
homogenized model) pileup undergoes a hot moulding proc-
ess, which delivers the final shape: as a
result of it the part presents several are-
as where the core has been squeezed be-
tween the skins, but also many local
skin curvatures, so that the component
presents a non uniform thickness. Con-
cerning the modelling of the core, it has
been meshed with 3D linear tetra ele-
ments in order to adapt to the skin ge-
ometry: this has lead to an FE model
having 17.000 nodes. The model was di-
vided into two parts: one zone where
the thickness variation could be consid-
ered negligible, and a region where a
high compression ratio of the core was
evident.
Both these parts inherited the stiff-
ness equivalent macro characteristics,
Figure 4: Detail of a trunk load floor section with honeycomb cores using the proposed methodology; the
only difference was on the equivalent
density given to these parts. For the nu-
merical/experimental validation, com-
ly 1.5 min. The results show that not formed over the simulated FRFs, by sum- parison the forced response computed
only is the deformation profile of the ming up the contributions of all the with the FE model just described, with
full model reproduced correctly by the nodes belonging to the skin opposite to the spectra on two load floor samples
homogenous one, but the local defor- the one where the loads are applied. Sev- are measured. These were suspended
mation is also correctly represented eral excitation points have been consid- with rubber bands on the upper cor-
leading to a maximum 3.5 % difference ered. Figure 3 shows the results where one ners, acquiring the vibrations on one
in the displacement field of these two of them is applied. floor skin by means of a laser scanner
models, Figure 2. Also in this behaviour comparison, on 221 points, Figure 5.
A further comparison between the despite the difference for both the mod- Figure 6 shows how well the proposed
two models involved the forced response elling effort and solution time (approxi- modelling can correlate to the measure-
of the sandwich panel, when exciting mately 6 h for the validation model and ments. Despite the high geometrical and
the sandwich panel, in free-free condi- 43 min for the homogenized model), the material complexity, the numerical mod-
tion, on one point with a force normal to proposed homogenized model is able to el can find a pretty good agreement with
the skin surface. The average mobility provide a response, which is pretty simi- measurements up to nearly 500/600 Hz.
frequency response function is per- lar to the one of the validation model. Only at higher frequencies does it seem

ATZ 01I2008 Volume 110 29


Development Lightweight Design

phase of honeycomb sandwich car parts


like for trunk load floor panels. It can
seamlessly handle parts with variable
core thickness and curved surfaces, re-
quiring a reduced FE meshing effort.
Moreover the honeycomb core solu-
tion can be changed either by varying
parametrically the core shape or by
changing the core material, without hav-
ing to re-mesh the FE component. There-
fore it can be applied in the design of
honeycomb sandwich and in the identi-
fication process of efficient cellular core
layouts with respect to NVH targets.

References
[1] Shi, G., Tong, P.: The Derivation of Equivalent
Constitutive Equation of Honeycomb Structures by
a Two Scale Method. In: Computational Mechanics
15 (1995), PP 395407
[2] Guj, L., Sestieri, A.: Dynamic modeling of honey-
Figure 5: Snapshot record with a laser scanner of a load floor deformation mode comb sandwich panel. In: Archive of Applied
Mechanics 77 (2007), No 11, pp 779793

Figure 6: Numerical/experimental comparison of root mean square (RMS) values for mobility
velocity

to lose its accuracy even though the dict the equivalent elastic tensor of hon-
trend of the experimental behaviour ap- eycomb cores with any general core con-
pears to be respected. figurations.
The numerical approach has been
validated using industrial finite ele-
5 Outlook ment software, onto several sandwich
constructions having various core con-
The work of FE simulation done by Rieter figurations, with respect to static and
in the past years has shown that the pro- dynamic analysis types. This simulation
posed homogenization approach can be has shown a great accuracy, as well as
efficiently used. It is thus possible to pre- direct implementation in the design

30 ATZ 01I2008 Volume 110